Murf AI plans to release an updated version of its Falcon conversational voice model within three weeks, according to Ankur Edkie, Co-Founder and CEO. The move could the company a sharper competitive edge as it looks to capture a larger share of the rapidly expanding global voice-AI market.

The launch follows Falcon 2, which executives said ranks ahead of OpenAI Realtime API and ElevenLabs’ Flash and Turbo models for naturalness on the Artificial Analysis Speech Arena. Edkie told Fe that Murf is focused on improving the speed and efficiency of its voice models given the importance of low latency in making AI-led conversations feel natural.

“What we need to look at, is latency being sub 200 milliseconds, at least. Beyond that, it’s not really usable. Because if you make a larger model, you’ll get better realism, but you lose on latency, Edkie said.

“There’s a trade-off between latency, price and realism. Murf’s models run four to five times faster than some rivals,” he said.

Falcon 2 is currently priced at $0.01 per minute of generated audio, compared to $0.05 for both ElevenLabs Turbo and Flash.

Murf, which began six years ago as an AI voiceover tool called Murf Studio before shifting toward enterprise conversational AI, said the U.S. remains its largest market but that India has emerged as a major growth opportunity. Its Indian client base is concentrated in the private sector– customer support, outbound calling, sales, and e-commerce and logistics.

Some of Murf’s largest e-commerce customers are currently routing 10-20% of their traffic through Murf’s API, according to Edkie.

Edkie said he does not consider language-model builders such as Sarvam AI, which are building broader, larger-parameter foundation models, as direct competitors. “It’s almost counterintuitive for a conversational AI system to have too many parameters,” he said, adding that Murf’s approach is to match the naturalness of larger models at a smaller parameter count to preserve speed.

Murf is likely to raise funds towards the end of 2026. The company’s current model supports 150+ voices across 35+ languages, including in-depth coverage of Indian languages.